Anthropic opens its most powerful model to the public: Claude Fable 5

Anthropic launched Claude Fable 5 on June 9, 2026 — the first Mythos-class model it has made available to the general public. 1

Mythos-class models sit above Anthropic's Opus tier in capability. The first, Claude Mythos Preview, went out in April restricted to a handful of cybersecurity partners due to concerns about its hacking abilities. Fable 5 is the same class of model, sold to anyone, with hard safety limits that block responses in areas Anthropic considers high-risk.

On Cognition's FrontierCode benchmark, which tests models on complex production-quality coding tasks, Fable 5 scores highest among frontier models. 1 Stripe reported in early testing that the model compressed months of engineering work into days: on a 50-million-line Ruby codebase, it ran a codebase-wide migration in a day that a full team would have taken over two months to complete by hand.

Knowledge work is comparably strong. On Hebbia's Finance Benchmark for senior-level reasoning, Fable 5 scored highest of any model tested. IMC, the trading firm, said it "aced their trading-analysis evaluations nearly across the board." Analytics company Hex said it was the first model to break 90% on its core analytics benchmark of complex, long-running tasks.

Vision tasks improved substantially. Fable 5 can extract precise numbers from dense scientific figures and rebuild a web app's source code from screenshots alone. Earlier Claude models needed a complex helper harness to play Pokémon FireRed; Fable 5 completed it with a minimal, vision-only harness.

The model has a 1-million-token context window and uses file-based memory to maintain focus across long autonomous runs. In testing on Slay the Spire, giving the model access to persistent memory improved its performance three times more than the same setup improved Opus 4.8's.

Anthropic's concern is straightforward: the same capabilities that make Fable 5 useful for finding software bugs make it useful for staging cyberattacks. Without controls, the model could provide meaningful uplift to malicious actors.

The solution is a layer of classifiers — separate AI systems that detect potentially dangerous queries before they reach Fable 5 and route them instead to Claude Opus 4.8. The trigger areas are cybersecurity, biology and chemistry, and distillation (large-scale extraction of Claude's capabilities to train competing models). 12

Anthropic stress-tested these classifiers before launch. An external bug bounty produced no universal jailbreaks in over 1,000 hours of testing. One of Anthropic's external partners found that Fable 5 complied with zero harmful single-turn requests relating to planning a cyberattack, exploit development, or defense evasion — holding across 30 different public jailbreak techniques.

The classifiers are tuned conservatively, which means they catch some harmless requests. Early data shows fewer than 5% of sessions involve a fallback to Opus 4.8, so in practice most users get Fable 5 responses almost all the time.

One side effect: Anthropic is requiring 30-day data retention on all Fable 5 and Mythos 5 traffic, including enterprises that previously had zero-retention agreements. The data won't be used for model training; the stated purpose is detecting novel jailbreak attempts and reducing false positives. 2

Alongside Fable 5, Anthropic simultaneously launched Claude Mythos 5 — the identical underlying model with the cyber safeguards lifted in some areas. 1 Access is restricted to organizations already in Project Glasswing, Anthropic's cybersecurity partnership with the US government. Those users are upgrading from Mythos Preview to Mythos 5 at roughly the same or stronger capability at less than half the prior price. Anthropic says it plans a broader trusted access program for cybersecurity organizations, and a separate biology-focused trusted access program allowing biomedical researchers to use Fable 5 with the biology and chemistry classifiers removed.

Both Fable 5 and Mythos 5 are priced at $10 per million input tokens and $50 per million output tokens — less than half the price of Claude Mythos Preview, and double the price of Opus 4.8. 3

Access on subscription plans is rolling out in stages. Through June 22, Fable 5 is included in Pro, Max, Team, and seat-based Enterprise plans at no extra charge. On June 23, Anthropic pulls it from those plans and access will require usage credits. The company says it intends to restore Fable 5 as a standard subscription feature as quickly as capacity allows.

Via the Claude API and consumption-based Enterprise plans, Fable 5 is fully available today.

Anthropic filed confidentially for an IPO on June 1 at close to a $1 trillion valuation, coming off a reported revenue run rate of $47 billion. 45 Fable 5 at $50 per million output tokens is twice as expensive as Opus 4.8, and customer testimonials from Stripe, GitHub, and Cursor suggest enterprises are willing to pay the premium for gains in autonomous task completion.

OpenAI filed its own confidential S-1 on June 8, setting up a direct IPO race. 2 Fable 5 is Anthropic's first product launch timed, in effect, as an IPO revenue story. The model converts what was a restricted government partnership program into a commercial product line. The question for investors will be whether $50/M output token pricing holds as competitors respond and whether the mandatory data retention policy creates friction with enterprise security buyers.